Active 3-D vision on a humanoid head

In this paper we describe and experimentally evaluate how to model, control, and use the capabilities of a humanoid visual system with foveated vision. We present a computational process that can be utilized to identify and update the parameters of the robot´s eyes under motion, which enables the use of 3D vision on an active humanoid head. We also derive the formulas expressing the geometry of our foveated vision setup. Based on these results we can actively control the eye gaze towards the potential regions of interest and analyze these areas using foveation and 3D vision processing. Experimental results showing the accuracy of the system are provided. The system has been demonstrated to be sufficiently accurate to realize grasping using active 3D vision.